2.Cartoon filter - Image filtering

Image filtering

1. What is image filtering

• Image filtering
  • The task filter the unecessary information and pass the necessary information to image

• Frequency domain filtering

• Spatial domain filtering
  • The method directly use the fixel value of video
  • Generally Using the mask operation (mask = kernel = window = template)
• The form and value of mask makes the task to define
  • Making smoothly image
  • Making sharply image
  • Edge detection
  • Noise reduction

• Example spatial domain filtering

• Basic 2D filtering

cv2.filter2D(src, ddepth, kernel, dst=None, anchor=None, delta=None, borderType=None) -> dst

• src : Input video
• ddepth : Data type of output image ex) cv2.CV_8U, cv2.CV_32F
• kernel : Filter mask metrix
• anchor : Position of fix point
• delta : Additional added value
• borderType : Border pixel Expension method
• dst : Output image

2. Blurring

• Mean filter
  • Specific coordination value of image set the arithmetical average of near fixel values.
  • The scale value change of each pixel decreases and the sharp edge is going to be smooth. so The effect of noise disappears.

• Mean filtering function

cv2.blur(src, ksize, dst=None, anchor=None, borderType=None) -> dst

• src : Input video
• ksize : Mean filter size ( tuple type - (width, height) )
• dst : Output video

• Mean filter example

3. Gaussian filter

• Disadvantage of the bluring using Mean filter
  • From the position of target, Both near pixel and distant pixel use the same weight for calculating the average.
  • The distant pixel could be affected

• 1-Dimension Gaussian function

• 2-Dimension Gaussian function

• Gaussian filtering function

cv2.GaussianBlur(src, ksize, sigmaX, dst=None, sigmaY=None, borderType=None) -> dst

• src : Input image
• dst : Output image
• ksize : Gaussian kernel size
• sigmaX : x-directional sigma
• sigmaY : y-directional sigma
• borderType : Edge pixel expension method

• Gaussian filter mask example

4. Sharpening

• Unsharp mask filtering
  • Unsharp image, Being the sharped image use to make sharp image.

• Realize unsharp mask filter

5. Reduction noise - median filter

• Noise of image
  • The Unexpected form of the signal be added at pixel value of the image.

• Sort of noise
  • Gaussian noise
  • Salt & Pepper

• Median filter
  • It replace pixel values by sorting to median value, where are near center position.
  • It is effect to reduce the Salt & Pepper noise

• Median filter function

cv2.medianBlur(src, ksize, dst=None) -> dst

• src : Input image
• ksize : Kernel size
• dst : Output image

• Median filtering example

6. Bilateral filter

• It’s one of edge-preserving noise removal filter
• It have weakness that Median filter and gaussian filter make to average out the pixel value nearby edge.

• Normal gaussian filtering : Blurring in the whole image.

• Bilateral filter : Burring in the out of edge

• Bilateral filtering function

cv2.bilateralFilter(src, d, sigmaColor, sigmaSpace, dst=None, borderType=None) -> dst

• src : Input image
• d : Distant(Diameter) of pixel
• sigmaColor : Standard deviation in color space
• sigmaSpace : Standard deviation in coordination space
• dst : Output image
• borderType : Edge pixel processing method